Omni Core · UI and UX Optimization · Module

Testing changes

Establishing whether a change improved anything, on a measure agreed beforehand, with enough volume to be readable.

The idea

How it works

The honest finding of most testing programmes is that the majority of changes have no measurable effect. Knowing that changes how much confidence a shipped change should carry.

Underpowered tests are worse than no tests, because they produce a result that looks like evidence. A test that cannot detect the effect size that matters should not be run.

Working with it

In practice

  1. 01

    Agree the measure before building

    And the size of change that would count. Deciding afterwards means the winning metric is chosen from the results.

  2. 02

    Check you have the volume

    If the traffic cannot detect a meaningful effect, the test will produce noise presented as a finding.

  3. 03

    Run it to the planned end

    Stopping when the result looks good is the most common way a test is invalidated.

  4. 04

    Record the null results

    They are the majority and they are what keeps expectations calibrated.

One level in

The components of testing changes

A component is something that exists afterwards which did not exist before — a deliverable or a mechanism, not an intention.

  1. The measure

    What the change is meant to improve, agreed in advance.

    Learn
  2. The power

    Whether the available volume can detect the effect that matters.

    Learn
  3. The result

    What happened, including the nulls.

    Learn

A test that cannot detect the effect size that matters should not be run. It will produce noise presented as evidence.